s4:dsdb:mod: repl_md: make message_sort transitive

Before we had (with a TODO of regret):

       if (!a1 || !a2) {
               return strcasecmp(e1->name, e2->name);
       }

so, given {name:"A", id 2}, {name:"B", NO id}, {name:"C", id 1},

 A < B by name
 B < C by name
 A > C by id

Now the sort order is always A > C > B.

This sort could have caused mysterious crashes in repl_meta_data if
the schema is out of sync.

BUG: https://bugzilla.samba.org/show_bug.cgi?id=15625

Signed-off-by: Douglas Bagnall <douglas.bagnall@catalyst.net.nz>
Reviewed-by: Andrew Bartlett <abartlet@samba.org>

diff --git a/source4/dsdb/samdb/ldb_modules/repl_meta_data.c b/source4/dsdb/samdb/ldb_modules/repl_meta_data.c
index 57098f0a8b789da51038aba9f01de1eebae4728a..ea7e8a0d4a4ce96b2e821d589c06a621621599aa 100644 (file)
--- a/source4/dsdb/samdb/ldb_modules/repl_meta_data.c
+++ b/source4/dsdb/samdb/ldb_modules/repl_meta_data.c
@@ -1063,12 +1063,20 @@ static int replmd_ldb_message_element_attid_sort(const struct ldb_message_elemen
        a2 = dsdb_attribute_by_lDAPDisplayName(schema, e2->name);
 
        /*
-        * TODO: remove this check, we should rely on e1 and e2 having valid attribute names
-        *       in the schema
+        * If the elements do not have valid attribute names in the schema
+        * (which we would prefer to think can't happen), we need to sort them
+        * somehow. The current strategy is to put them at the end, sorted by
+        * attribute name.
         */
-       if (!a1 || !a2) {
+       if (a1 == NULL && a2 == NULL) {
                return strcasecmp(e1->name, e2->name);
        }
+       if (a1 == NULL) {
+               return 1;
+       }
+       if (a2 == NULL) {
+               return -1;
+       }
        if (a1->attributeID_id == a2->attributeID_id) {
                return 0;
        }